1. Introduction
Global digitization has led to exponential growth in digital information [
1]. However, the density and long-term stability of silicon-based storage technologies are approaching their theoretical limits [
2], [
3]. Moreover, the manufacturing process of silicon-based storage media generates large quantities of environmental hazards, including the media themselves [
2], [
3], [
4], [
5], which is not consistent with the global vision of green and sustainable development [
6]. A dense, long-term stable, and sustainable storage technology is highly desirable to meet the future needs of digital data storage [
2], [
7], [
8], [
9], [
10], [
11], [
12]. DNA is a naturally formed polymer carrying the genetic information of all life on earth [
13], [
14]. Studies have revealed that DNA can also be utilized for the storage of massive amounts of artificial information [
7], [
15], [
16], [
17], [
18], [
19], [
20], [
21], [
22], [
23], [
24], [
25], [
26], [
27], [
28]. DNA data storage utilizes the complex processes of DNA synthesis and sequencing for the “writing” and “reading” of information. Uniquely, data-encoding DNA molecules can be mixed together, forming a natural three-dimensional (3D) storage system. The extra dimension allows DNA data storage with the outstanding feature of ultra-high density, outpacing traditional planner media [
9], [
19], [
28], [
29], [
30], [
31], [
32], [
33]. While recent studies have made significant progress in terms of data scale, stability, random access, data replication, and so on, cost has emerged as the key barrier to a practical DNA data-storage system [
7], [
15], [
16], [
17], [
18], [
19], [
20], [
21], [
33], [
34]. The reading cost of DNA data storage is relatively high compared with that of traditional mediums [
14], [
20]. However, the cost of reading technology is not a significant limitation to the practical applications of DNA data storage, considering the low reading frequency in its main application scenario of long-term archive storage. The main obstacle at present is the expensive and environmentally unfriendly data-writing process of chemical-based DNA synthesis. It has been estimated that a decrease of five to six orders of magnitude is required in the writing cost in order to outpace magnetic-tape-based storage technology [
2], [
9], [
14], [
20].
Movable-type printing is the crystallization of the wisdom of ancient China. It uses pre-made movable types rather than fixed-order printing plates, which makes it possible to print any type of documents by placing prepared movable types in a certain order during the printing process [
35]. This strategy greatly reduces the cost of text printing through the repetitive and flexible usage of movable types. In this study, inspired by movable-type printing technology, we designed and implemented a movable-type-like system for DNA data storage. This DNA movable-type storage system utilizes natural pre-synthesized DNA fragments as basic writing units—that is, as DNA movable types. Rather than depending on DNA synthesis, the data-writing process is achieved by the selection and assembly of specific DNA movable types into longer storage units. A fast and reliable enzymatic assembly process has been designed and optimized for these short fragments, ensuring reliable data writing in DNA. This system, through its repetitive usage of pre-synthesized DNA movable types, shows tremendous promise for low-cost data writing. With further advancements, it has the potential to address the high-cost issue of current DNA data-storage technologies. Significantly, as far as our knowledge extends, this system is a pioneering effort in the realm of data writing, as it is the first system reported that can utilize natural DNA molecules produced by cell factories. This opens up a novel and sustainable approach to digital data storage.
2. Material and methods
2.1. Construction of DNA movable types
All polymerase chain reaction (PCR) amplifications were performed with Phanta Max Super-Fidelity DNA Polymerase (CAT#: P505-d2; Vazyme Biotech Co., Ltd., China) in a 50 µL volume. For the initial construction of each DNA movable type, a 45-base pair (bp) single strand DNA (ssDNA) fragment was used as templates for the PCR amplification. The 6 bp data encoding region is located in the center of the ssDNA fragment. The PCR amplifications were performed with the following primes: P1: 5′-GTCGGTCTCGTCAGATGTAGATAACTACGACGCTCGTGATCATCCATTCTCGTCTAC-3′; P2: 5′-GTGAAGACTCCTGAGTCATTCTCAGAATGACGTAGTTGGTGGTGGAACCTGAGAAGC-3′. The two primers were used to introduce two restriction sites of BsaI and BbsI. The resulting DNA movable-type fragment is 120 bp in length. The PCR amplifications were performed using the following program: an initial denaturation step at 95 °C for 30 s, followed by ten cycles of 95 °C for 15 s, 48 °C for 15 s, and 72 °C for 4 s. This was followed by 20 cycles of 95 °C for 15 s, 68 °C for 15 s, and 72 °C for 4 s, and a final extension step of 72 °C for 5 min, with a final hold at 4 °C.
To prepare a large quantity of DNA movable-type fragments, a different pair of primers with a shorter length was used, as follows: P1: 5′-GTCGGTCTCGTCAGATGTAGATAACTACGAC-3′; P2: 5′-GTGAAGACTCCTGAGTCATTCTCAGAATGAC-3′. The applied PCR amplification program was as follows: 95 °C for 30 s, 95 °C for 15 s, 59 °C for 15 s, and 72 °C for 4 s for 30 cycles; followed by 72 °C for 5 min and hold at 4 °C.
The PCR products were purified by gel purifications before usage. All the gel purifications were performed using the SPARKeasy Gel/DNA Extraction Kit (CAT#: AE0301-B; Shandong Sparkjade Biotechnology Co., Ltd., China).
2.2. Assembly of DNA movable types
The DNA movable types were digested with fast-digesting enzymes before assembly. Thermo Fisher’s FastDigest series enzymes (CAT#: FD0294, FD1014) were used for all the restriction enzyme digestion experiments. To perform the digestion, 27.5 µL of the DNA movable-type solution, 5 µL of 10× buffer, and 1 µL of specific restriction enzyme were mixed in a tube and incubated at 37 °C for a controlled time ranging from 5 to 30 min. The digested DNA movable-type fragments were purified before assembly.
For the ligase-mediated assembly, 11 µL of each DNA movable-type fragment, 3 µL of 10× buffer, and 1.5 µL of T4 DNA ligase (CAT#: M0202L; New England Biolabs (NEB), USA) were mixed in a tube, double distilled water (ddH2O) was also added to make a total volume of 30 µL. After ligation, the mixture was inactivated by incubating it at 65 °C for 10 min, according to the manual.
In the series of experiments to optimize the ligation process, the same ligation system was used as above, except for the dosages of the ligase and ligation time. Ligase dosages of 0.50, 0.75, 1.25, and 1.50 µL were tested with a ligation time of 20 min. The resulting ligation samples were analyzed using the Agilent Bioanalyzer 2100 system (Agilent Technologies, Inc., USA).
2.3. Cost-estimation settings
In the biological preparation of the DNA movable types, plasmid extraction was performed using the TIANprep Mini Plasmid Kit (CAT#: DP103; Tiangen Biotech (Beijing) Co., Ltd., China). EcoRI-HF was used to digest the extracted plasmid, and the buffer used was CutSmart Buffer from NEB (CAT#: R3101L).
2.4. Data and materials availability
All the original algorithms proposed in this study were implemented with Python; the source codes are provided in Appendix A. All sequencing data can be retrieved at
https://doi.org/10.6084/m9.figshare.19425236.3. Results
3.1. The design of the DNA movable-type storage system
As illustrated in
Fig. 1(a), DNA movable types with a specific length of data-encoding region are used as basic writing units. Each specific DNA movable type contains a unique sequence combination in the encoding region. All the DNA movable types of all possible sequence combinations must be pre-produced before the actual writing operations can take place. This makes it possible to write arbitrary digital information via the automatic selection and ordered assembly of specific DNA movable types. Importantly, these DNA movable-type fragments are utilized repetitively in this system. The initial synthesis of the DNA movable-type fragments adopts mature oligonucleotide chemical synthesis methods, and subsequent large-scale production can be carried out through biological manufacturing and separation. For data encoding, specific binary information is first transcoded into DNA strings according to the mapping schema shown in
Table 1. To avoid decoding failures, one error-checking base is inserted for every two bases, using a strategy reported in a previous study [
29]. The DNA sequence generated is then split into massive storage units. Next, to realize the writing of data, a high-throughput device is employed to pick the corresponding DNA movable types and produce the required storage units by means of an optimized process of fragment assembly, as illustrated in
Fig. 1(b).
As the core module of the DNA movable-type storage system, a unified structure of the DNA movable types is crucial. As shown in
Fig. 1(c), to support the assembly of two arbitrary DNA movable types with a defined order, two Type IIS restriction sites are respectively placed on the two ends of the DNA movable types. BsaI and BbsI are Type IIS endonucleases with the key feature of detached recognition and cleavage sites. With this important feature, the recognition sites can be eliminated after the enzymatic digestion by designing appropriate DNA movable types, thereby avoiding mis-cleavage of the assembled DNA movable types in the next round of assembly. In addition, the issue of self-ligation of DNA movable types—a common issue with conventional Type II endonucleases that possess a palindromic sequence in the cleavage site—can be avoided. The two restriction sites have been designed to produce compatible sticky ends, enabling the assembly of two arbitrary DNA movable types via selective digestion of the two ends with the corresponding restriction enzyme. To avoid strand dissociation of the double strands of the DNA movable types, two helper fragments are added to increase the melting temperatures. This also increases the ligation efficiency of T
4 ligase, which is more efficient with double-stranded DNA (dsDNA). Finally, a basic fragment with a length of 120 bp is formed, as shown in
Fig. 1(d) and Fig. S1 in Appendix A. The fragments—that is, the DNA movable types—are assembled into storage units after specific rounds of assembly to accomplish the data-writing process. All the storage units can be collected and stored at low temperature for long-term storage. To read the stored data, the storage units must be sequenced by an appropriate sequencing device, such as an Illumina sequencer.
The storage capacity of a system represents its storage potential, and whether it can flexibly cope with the data size is crucial for large-scale data storage. For storage units with a length of
L in which the index length is
x, the length of the data-encoding region is
L −
x. The information-storage capacity of a single base pair is 2 bits. Thus, the maximum total amount of information that can theoretically be stored in storage units with a length of
L is
N = 4
x × (
L −
x)
× 2 bits. It can be concluded that the storage capacity of the fragments increases with an increase in the index length; that is, when
x =
L − 1,
N has the maximum value. Based on this formula, with storage units that are 24 bp long, the theoretical storage capacity of this system can reach about 17 terabytes (TB; 1 TB = 10
12 bytes). In practice, the error correction (EC) codes should be excluded from the calculation. With the three-base block coding applied, the effective coding length of each storage unit is 16 bp (24/3
× 2 = 16), and a minimal data-encoding region is two bases. Thus, we obtain a storage capacity of 134 megabytes (MB; 1 MB = 10
6 bytes). Importantly, the storage capacity of this system can be sustainably improved by using longer storage units, which can be achieved using either DNA movable types with a longer encoding region or more rounds of assembly, as shown in
Table 2.3.2. Design and optimization of the assembly process
The DNA movable-type fragments were digested and then the storage units were assembled via ligation-based assembly, using T
4 ligase. T
4 ligase is very sensitive to temperature. To optimize the assembly efficiency and shorten the assembly time, we tested the ligation efficiency of T
4 ligase at various reaction times (15 s, 30 s, 1 min, 2 min, 5 min, and 10 min) and temperatures (16, 26, and 37 °C).
Figs. 2(a) and (b) show that ligation products can be observed after a 15 s reaction time, with the efficiency of ligation steadily improving with increased ligation time. In the experimental group, ligase at 16 °C showed good ligation efficiency within 2 min, making it suitable for cases with a tight assembly time. When the reaction is carried out at 26 °C, the ligation efficiency shows a more significant growth trend with the extension of time, particularly in the 5-10 min interval, which exceeds the effect at 16 °C for the same time. At this temperature, the concentration of assembled fragments is relatively high, allowing direct support for subsequent experiments without amplification. Moreover, there is no need for temperature adjustment facilities like a PCR instrument, making it easy to operate under normal environmental requirements. However, when the ambient temperature is changed to 37 °C, the ligation effects only slightly increase with prolonged ligation time. After incubation at 37 °C for more than 5 min, the effect is comparable to that of 1 min at 16 °C. When the temperature is increased to 45 °C, the ligation efficiency no longer shows a significant trend with time, but ligated fragments can still be observed, indicating that a certain ligation effect was achieved. Although the concentration of ligated fragment is low, they could be subsequently amplified using PCR before the next round of assembly.
These results indicate that reliable data writing can be achieved in the DNA movable-type storage system under normal operating temperatures ranging from 16 to 45 °C. The main factor contributing to the writing cost is the consumption of ligase. Therefore, reducing the amount of ligase used in the system could decrease the writing costs. In an additional experiment, the ligase added was gradually reduced from 1 unit to 5/6, 2/3, 1/2, and 1/3 of a unit at room temperature (26 °C). Although a slight decrease in ligation efficiency was observed as the ligase dosage was reduced, the subsequent assembly process remained reliable (
Figs. 2(c) and (d)). This assembly-based writing process is highly adaptable and allows for balancing the cost and action time by adjusting the dosage of added enzymes and the reaction time.
3.3. Proof-of-concept verification using text information
A short text with the first names of three scientists who made great contributions to the discovery of the DNA double-helix structure—namely, “Watson, Crick, Franklin.”—was used as the input information for the proof-of-concept verification. To deal with the base error of data in DNA synthesis, storage, amplification, and sequencing and to ensure the reliable recovery of information, block EC codes were introduced in the process of binary data conversion. Finally, a string of base sequences with a length of 144 bp was obtained. This sequence was split into a series of small fragments of 6 bp for a total of 24 DNA movable types (Table S1 in Appendix A). These DNA movable types were then found in the pre-generated fragment “pool” by means of location screening. Each DNA movable type is 120 bp in length, as shown in
Fig. 1(d). Four DNA movable types were assembled into one storage unit after two rounds of assembly (
Fig. 3(a)). For each assembly process, 24 nucleic acid residues were removed. Thus, the length of an assembled storage unit is 408 bp (120 × 4 - 24 × 3 = 408). As shown in
Fig. 3(b), agarose gel electrophoresis analysis of the assembled products revealed a band corresponding to a length of 250-500 bp, as expected. To reduce the space occupied, the fragments can be mixed together. However, the storage units are out of order in this case. In order to decode the data correctly, the order information of the storage units must be encoded in a certain way. To achieve this, the first two DNA movable types of each storage unit were designed to be the same as the last two DNA movable types of the previous storage units, to determine the order of the storage units (
Fig. 3(c)). A total of 11 storage units were assembled to encode the 24 bytes of text information. It should be noted that a small fraction of the storage unit fragments may be lost without the possibility of recovery when massive storage units are utilized for large data storage. Thus, additional erase codes are definitely required to handle this issue [
7], [
15], [
16], [
17], [
18], [
19], [
36]. Data retrievals were performed independently with Sanger sequencing and Illumina sequencing methods. All storage units encoding tiny pieces of information were mixed in a tube for high-throughput sequencing via the Illumina sequencing platform. In both cases, perfect data recovery was achieved with the obtained sequencing reads (Figs. S2-S13 and Table S2 in Appendix A), demonstrating the feasibility of the proposed DNA movable-type storage system.
3.4. Estimation of the potential of low-cost data writing
The cost of this storage system is comprised of the production cost of the DNA movable types and the assembly cost for generating the storage units. The production cost of the DNA movable types has two parts: the initial construction cost and the reproduction cost. For the initial construction of all DNA movable types, the synthesis cost of each base in the DNA single strand is around 0.039 USD, based on previous reports [
14], [
20], and the length of each single-strand template is 90 bp. For the 6 bp data-encoding region, there are 4096 (4
6 = 4096) different DNA movable types, considering all the sequence combinations. The two universal primers are both 31 bp in length. With an optimized production process, the required synthesized single-stranded template length was reduced to 45 bp and both universal primers are 57 bp in length, resulting in a synthesis cost of 7192.926 USD (0.039 × (45 × 4096 + 57 × 2) = 7192.926).
It should be clarified that this initial manufacturing process only needs to be performed once. In other words, this initial construction cost is the research and development (R&D) cost, not the data-writing cost in practical usage. The reproduction of DNA movable types can be achieved via cheap PCR or even cell factories (Fig. S14 in Appendix A). Importantly, the DNA movable types can be used multiple times for the assembly of various storage units—that is, for the storage of various digital data. As shown in this study, even at a low dosage of 0.5 ng⋅µL−1, efficient assembly can still be achieved. To achieve economical and accurate production of DNA movable-type fragments, large-scale production can be performed using cell factories, which can further reduce the writing costs. Considering that the commercial price of the kits used includes considerable profit and R&D recovery funds, and the cost of large-scale production would be much lower than this price. Therefore, 1% of the commercial price of the kits is used to calculate the cost more reasonably.
Assuming the volume of a single tube for cell cultivation is V1, and take V1 = 5 mL for the cost estimation. The main cost of the cell cultivation comes from the medium. Taking the price of the lysogeny broth (LB) medium for estimation, a1 = 0.63 USD∙L-1 (63.16 × 1% ≈ 0.63), the cost of LB medium of 5 mL cell culture is W1 = a1V1/1000 = 3.15 × 10−3 USD. According to the plasmid extraction kit instructions, a maximum of 5 mL of plasmids can be extracted at one time. According to the kit specifications and price, the cost of a single plasmid extraction is a2 = 5.7 × 10−3 USD (0.57 × 1% = 5.7 × 10−3), and the cost of a single test tube plasmid extraction is W2 = a2 = 5.7 × 10−3 USD. For the high-copy plasmid pUC19, a plasmid yield of 5 mL of bacterial liquid can reach m1 = 30 μg, and the concentration of plasmid in the final extract is c = 6 µg·mL−1. The main cost of enzyme digestion and product recovery comes from the reagents used in the enzyme digestion process. According to the price and dosage of the reagents used in digesting the extracted plasmid, the digestion cost is approximately a3 = 0.79 USD·mg−1 (78.8 × 1% ≈ 0.79) of DNA, and the cost of the digestion and product-recovery steps is approximately W3 = a3m1/1000 = 2.37 × 10−2 USD. In this experiment, the OD600 (OD600 stands for the optical density at 600 nm) of the bacterial solution reached 2.4 after overnight cultivation, corresponding to a bacterial solution concentration of about C = 2.4 × 108 mL-1, while the copy number of the pUC19 plasmid was about e = 600. Assuming that the extraction efficiency of the short film segment η = 50%, then the number of short fragments of information-storage DNA that can be extracted from 5 mL of bacterial solution is n = C·V1·e·η = 3.6 × 1011. According to the results of the above calculation, the total cost of the production of short DNA fragments is approximately W = W1 + W2 + W3 ≈ 3.26 × 10−2 USD. The effective information-storage capacity of one storage unit in this system is S = 6 bytes; for each information-storage instance, the number of copies requiring the same DNA short fragments n1 ≈ 1000. As such, the biological preparation cost of DNA movable types is W′ = n1W/(nS × 10−6) ≈ 1.5 × 10−5 USD·MB−1. This system has the potential to significantly reduce the writing cost of DNA data storage by several orders of magnitude.
4. Discussion and conclusions
DNA molecules are green materials that are abundant in nature. However, naturally produced DNA molecules cannot yet be used for the storage of artificial digital data. The reliance on expensive and hazardous DNA synthesis methods for data writing restricts the practical applications of current DNA data storage solutions [
37]. In this study, inspired by movable-type printing technology, we implemented a DNA movable-type storage system. With this system, pre-produced DNA materials can be engineered as basic writing units for the storage of arbitrary digital information. Through the repetitive usage of these basic writing units, herein referred to as “DNA movable types,” this system exhibits great potential for reducing the cost of data writing in DNA data storage. The data-writing process of this DNA movable-type storage system is achieved through rounds of assembly, producing the storage units with the desired length. As shown in
Table 2, the storage capacity of this DNA movable-type storage system can be strikingly enhanced by adding more rounds of assembly. As shown here, in the case of DNA movable types with an encoding length of 6 bp, the storage capacity of this system is as high as a staggering 576 petabytes (PB; 1 PB = 10
15 bytes) with merely three rounds of assembly. The storage capacity can be further enhanced by using DNA movable types with longer encoding regions. With encoding regions of 9 and 12 bp, the storage capacity could be increased to 2 brontobytes (BB; 1 BB = 10
27 bytes) and 11 corydonbytes (CB; 1 CB = 10
36 bytes), respectively, showcasing the system’s potential for large-scale data storage. Alternatively, by incorporating additional rounds of assembly, the storage capacity could be greatly boosted without the need for an increase in the total number of DNA movable types. It should be noted that using the highest coding capacity scheme will result in lower coding efficiency. Coding efficiency is considered to be essential as it impacts the cost and physical density—two crucial parameters indicating the advancement of a DNA storage system. Coding efficiency has gained great attention previously, especially in DNA synthesis-based methods where it directly affects the writing cost [
16], [
17], [
18]. However, in DNA movable-type storage system, although coding efficiency is still important, the cost and density are less sensitive to it. Theoretically, high-density and low-cost data storage can be achieved by reducing the strand copies of the DNA movable-type fragments, even with low coding efficiency. In practice, as long as high density and low cost are achieved, relatively low coding efficiency is no longer an issue. The total number of DNA movable types is a crucial parameter determining the complexity of implementing an automatic machine to execute the write operations. Thus, adding more rounds of assembly is the preferred strategy to increase the storage capacity of the system. Through process optimization, it takes less than 5 min to accomplish one round of assembly, suggesting a fast data-writing process compared with DNA synthesis-based data writing. Importantly, the production of these DNA materials and the assembly-based data-writing process being completely biological processes, make the current system a green and sustainable DNA data-storage technique. Furthermore, this data-writing process can be easily parallelized. With the ongoing advancement of high-throughput automation solutions, the data write bandwidth of this system is expected to be improved substantially. In conclusion, the DNA movable-type storage system presents a promising route towards economical, sustainable, and environmentally friendly data storage solutions to meet the future demands of large data storage.
the National Key Research and Development Program of China(2018YFA0900100)
the Natural Science Foundation of Tianjin, China(19JCJQJC63300)
PDF
(1305KB)
